Most companies bolt an assistant onto a database it cannot interpret. On BizBlocks, every entity carries machine-readable meaning, and an MCP server exposes that live schema so agents work against your real business safely.
Entities are attributed with display names, descriptions, business domain, relationships, validation rules and usage context. That metadata was built for developers and tooling, and it turns out to be exactly what an agent needs to write a correct query.
Administrators can curate it further from the UI, editing the descriptions that agents read.
Read the framework docs →Your live schema, including tables, views, procedures and functions, described to agents through the Model Context Protocol. Access is scoped and read oriented.
Each tenant can run its own assistants with their own system role, instructions, model, training sources and tool endpoints. Multiple bots per company, each with a purpose.
Train classification, regression, multiclass and anomaly models against your own tables, with configurable algorithms, versioning and saved model storage.
// One call, many providers, automatic fallback
var answer = await Llm.AskAsync(
prompt: "Summarize this month's exceptions",
model: "gpt-5");
// Result carries the telemetry you need
answer.Provider; // which one served it
answer.UsedFallback; // did we fail over
answer.TotalTokens; // what it cost
answer.LatencyMs;A gateway inside the framework routes requests across providers with automatic failover, smoothing over each vendor's quirks and returning consistent telemetry for provider, model, fallback, latency and token usage.
Providers are configuration, not code. When the frontier moves, your application does not have to.
Spin up a branded BizBlocks environment in seconds, or tell us how your business works and we will scope it with you.